Economics and Human Biology 44 (2022) 101089
Contents lists available at ScienceDirect
Economics and Human Biology
journal homepage: www.elsevier.com/locate/ehb
The origins of cognitive skills and non-cognitive skills: The long-term effect
of in-utero rainfall shocks in India☆
Grace Chang a, Marta Favara b, *, Rafael Novella c
a
London School of Economics and Political Science, Houghton St, Holborn, London WC2A 2AE, UK
University of Oxford, UK
c
University College London, UK
b
A R T I C L E I N F O
A B S T R A C T
JEL Classification:
J24
I14
Skills are an important predictor of labour, education, and wellbeing outcomes. Understanding the origins of
skills formation is important for reducing future inequalities. This paper analyses the effect of shocks in-utero on
human capital outcomes in childhood and adolescence in India. Combining historical rainfall data and longi­
tudinal data from Young Lives, we estimate the effect of rainfall shocks in-utero on cognitive and non-cognitive
skills development over the first 15 years of life. We find negative effects of rainfall shocks on receptive vo­
cabulary at age 5, and on mathematics and non-cognitive skills at age 15. The negative effects on cognitive skills
are driven by boys, while the effect for both cognitive and non-cognitive skills are driven by children of parents
with lower education, suggesting that prenatal shocks might exacerbate pre-existing inequalities. Our findings
support the implementation of policies aiming at reducing inequalities at very early stages in life.
Keywords:
Skills formation
In-utero
Rainfall shocks
India
1. Introduction
The foetal origins hypothesis (FOH) advocated by David J. Barker
proposes that the in-utero period is an important and critical period
where adverse (or favourable) conditions can have persistent and longterm effects on adult health (Barker, 1990, 1998). Since then, growing
economic literature finds that shocks that occur during the in-utero
period can affect various future outcomes such as adult health, human
capital, and earnings (Almond and Currie, 2011). Research in epide­
miology and developmental neuroscience suggests that the prenatal
period is crucial in influencing the brain structure and neural develop­
ment which subsequently affect cognitive function (Rooij et al., 2010;
Andersen, 2003; Thompson and Nelson, 2001). However, little is known
about the importance of this period for the formation of personality and
non-cognitive skills. Understanding the early formation of non-cognitive
skills is of particular interest given the impacts of these skills on key
economic outcomes later in life, such as employment and earnings
(Heckman et al., 2006; Cunha and Heckman, 2008; Cunha et al., 2010),
academic achievement, and social competence (Borghans et al., 2008;
Almlund et al., 2011).
This paper analyses the effect of shocks that occur in-utero on both
cognitive and non-cognitive skills development over childhood and
adolescence. The exposure to weather shocks among pregnant mothers
can affect their children human capital development through three po­
tential mechanisms. First, weather shocks can generate a dispropor­
tionate amount of stress and interfering directly on brain development in
a very critical period (biological channel). Second, weather shocks might
affect crops production and the related changes in prices and income
consequently would affect the consumption of food and other health
inputs (Hoddinott, 2006; Skoufias and Vinha, 2013) (nutrition channel).
Finally, the (net) impact of the shock will depend on the compensating
mechanisms intervening in mitigating the detrimental effect of the shock
with potential distributional effects.
This paper exploits rainfall fluctuations to test: (i) whether exposure
to environmental shocks during pregnancy negatively affects cognitive
and non-cognitive skills development throughout childhood and
adolescence; (ii) whether the effects vary depending on the intensity of
the shock and (iii) the number of shocks suffered during the gestation
period; (iv) whether the impacts of being exposed to shocks differ across
pregnancy trimesters, and (v) whether the in-utero shocks have a
distributional effect by gender and socio-economic status.
Our identification of the causal effect of rainfall variation on
We gratefully acknowledge excellent research assistance by Liza Benny and Kristine Briones. We also acknowledge comments by seminar participants at the
Centre for the Study of African Economies Conference 2019 and anonymous reviewers.
* Corresponding author.
E-mail addresses: g.c.chang@lse.ac.uk (G. Chang), marta.favara@qeh.ox.ac.uk (M. Favara), r.novella@ucl.ac.uk (R. Novella).
☆
https://doi.org/10.1016/j.ehb.2021.101089
Received 2 January 2021; Received in revised form 25 November 2021; Accepted 29 November 2021
Available online 2 December 2021
1570-677X/© 2021 Elsevier B.V. All rights reserved.
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
cognitive and non-cognitive skills development relies on the assumption
that, conditional on community-by-month fixed-effects, temporary
rainfall deviations from historical averages are uncorrelated with other
latent determinants of skills development during gestation and through
childhood and adolescence.
Our analysis uses the Young Lives (YL) data, a longitudinal dataset of
children born between January 2001 and June 2002 in Andhra Pradesh
(nowadays including the states of Andhra Pradesh and Telangana) and
followed for five rounds of data collection over 15 years. We combined
the YL dataset with monthly frequency gridded information on precip­
itation from the University of Delaware to construct a community-bymonth weather dataset that spans between 1900 and 2014. Andhra
Pradesh is vulnerable to several climate shocks including cyclones,
storm surges, floods, and droughts. According to the Revenue Disaster
Management of Government of Andhra Pradesh and UNICEF, multiple
incidences of heavy rain and flooding has been registered between April
2000 and September 2001, corresponding to the gestational period for
the YL children. Also, there were reports of drought in the first six
months of 2000 in the southern districts of Andhra Pradesh.1
Our study contributes to the literature on the effect of shocks in-utero
on long-term human capital development in several ways. First, this is
one of the few papers in the economic literature investigating the effect
of weather shocks happening during the gestation period. Second, we
investigate the effect on both cognitive and non-cognitive skills, where
evidence on the latter is particularly scarce. Third, we add to the thin
body of evidence on how effects evolve over time, throughout childhood
to adolescence. Finally, we contribute to the growing literature on the
long-term effects of more frequent aggregate shocks, which are far less
extreme compared to catastrophic shocks (such as famine episodes and
earthquakes) but affect larger populations and are likely to keep
occurring in the future.
We find that children who are exposed to rainfall shocks in-utero
have lower cognitive skills at age 5 and age 15. More specifically, we
find that being exposed to a rainfall shocks in-utero reduces the recep­
tive vocabulary test score (as measured by the Peabody Picture Vocab­
ulary Test - PPVT) at age 5 (in 0.15 points or 5% lower score than the
control group including children not affected by any shock in-utero) and
the math test score (in 13.6 points or 2% lower score respect to the
control group) at age 15. Additionally, rainfall shocks in-utero reduces
children’s core-self-evaluation (CSE), a composite measure of selfesteem, self-efficacy, and locus of control at age 15 by 0.16 points. No
statistically significant effects were found between ages 8 and 12. We
also document heterogeneity in our findings. The negative impact on
children’s cognitive scores were driven by boys, but there are no gender
differences for non-cognitive scores. Additionally, children of lower
educated parents are affected by the shock more than children of parents
who had at least completed primary education. Finally, according to our
results, being exposed to shocks after the first trimester of pregnancy has
the largest detrimental effects on children’s numeracy at age 15 and not
differential effect on PPVT scores and CSE scores.
The remainder of the paper is organized as follows. Section 2 briefly
reviews the existing literature on the effect of in-utero shocks on human
capital development and the potential transmission channels. Section 3
describes the two sets of data used in this paper and the main definitions
of the outcome variables of interest, how the gestational period and
rainfall shocks are defined. Section 4 shows some descriptive evidence
emerging from the data. Section 5 describes the empirical approach and
Sections 6 and 7 present and discuss the results and their validity. Sec­
tion 8 concludes with a summary and discussion.
2. Literature review
2.1. The effect of in-utero shocks on human capital development and later
outcomes
Economists have sought to establish the link between prenatal con­
ditions and human capital outcomes frequently exploiting natural
experiment to demonstrate causal pathways. Natural experiments either
in the form of climate shocks (Maccini and Yang, 2009; Kumar et al.,
2014; Andalon et al., 2016), pandemics (Almond, 2006; Banerjee et al.,
2010; Fletcher, 2018), famine (Neelsen and Stratmann, 2012), and
genocide (Bundervoet and Fransen, 2018) offer a suitable solution to the
omitted variables bias concern. One of the first economics study inves­
tigating the FOH was conducted by Almond (2006) who studied the
long-term effects of in-utero exposure to the 1918 influenza pandemic in
the US. He found that cohorts who were in-utero during the pandemic
displayed reduced educational attainment, increased rates of physical
disability, lower income, lower socioeconomic status, and higher
transfer payments compared with other birth cohorts. Similarly, Bun­
dervoet and Fransen (2018) found that children who were in-utero
during the genocide in Rwanda were approximately 8% less likely to
complete primary school and completed 0.3 years of education less than
children who were born a couple of months later.
While the early literature using natural experiments tend to focus on
disasters or more extreme shocks, there has been growing economic
literature investigating the effect of in-utero exposure to more frequent
aggregate events on future outcomes. For example, Almond et al. (2015)
and Almond and Mazumder (2011) find that Muslim students exposed to
Ramadan in the first half of pregnancy have respectively significantly
lower math test scores (between 0.06 and 0.08 standard deviations
lower) and are 20% more likely to be disabled as adults, the effect being
larger for mental (or learning) disabilities. Similarly, Majid (2015)
showed that children in Indonesia exposed to Ramadan in-utero scored
7.8% lower on cognitive tests and 5.9% lower on maths scores.
Similar evidence emerges from studies investigating the effect of
weather fluctuations during the gestation period on children’s health.
Rocha and Soares (2015) find that rainfall shocks during pregnancy can
lead to higher infant mortality, lower birthweight and shorter gestation
periods in Brazil. Andalon et al. (2016) find that in Colombia, in-utero
exposure to moderate low-temperature shocks during the first and sec­
ond trimester of pregnancy reduces children’s length at birth while
exposure to moderate heat waves in the third trimester reduces the
child’s birthweight. In India, Kumar et al. (2014) find that children who
experienced drought in-utero have poorer health, measured by
weight-for-age z-scores. Similarly, Ahmed and Ray (2017), using YL
data, find that children exposed to multiple shocks in-utero have lower
weight-for-age and height-for-age z-scores. Notably, the latter paper
measures shocks using self-reported information, which may suffer from
inaccuracies and recall bias. Our paper overcomes this problem by using
direct measures of rainfall data, which accurately measure weather
shocks experienced by the household.
There is a growing number of studies analysing the effects of weather
shocks on educational outcomes, mainly measured by educational
attainment and enrolment and a few on earning outcomes. Maccini and
Yang (2009) find that higher rainfall in-utero raises adult women’s
schooling and socio-economic status in rural Indonesia, but not for men.
Thai and Falaris (2014) find that negative rainfall shocks in-utero delays
Vietnamese children’s school entry and grade progression, between the
ages 6 and 19. In India, studies using data from ASER of primary school
children, find that children exposed to drought in-utero are less likely to
enrol in school, more likely to repeat a grade, and perform worse than
their peers in mathematics and reading tests (Shah and Steinberg, 2017).
1
For more information see: https://reliefweb.int/report/india/india-floodsappeal-no-192000-final-report; https://reliefweb.int/report/india/unicef-repo
rt-drought-and-floods-india-28-sep-2000
2
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
2.2. The effect of in-utero shocks on personality and non-cognitive skills
development
and consumption. This is particularly relevant in developing countries
where a large share of agriculture may be rain-fed and families live out
of what they produce, and where mitigation mechanisms are limited.
The effect on crops yields varies depending on the farmers’ choices in
selecting types of crops and productivity-enhancing inputs and their
ability to anticipate weather fluctuations (Amare et al., 2018; Bhavani
et al., 2017; Dillon et al., 2015; Palanisami et al., 2015). In the
short-term, the effect on consumption largely depends on the capacity of
the household to respond to the shock by smoothing consumption and
using savings. Based on existing literature (Hoddinott, 2006; Datar et al.,
2013; Rosales-Rueda, 2018; Skoufias and Vinha, 2012; Skoufias and
Vinha, 2013), we expect greater susceptibility of lower socio-economic
status households to adverse rainfall shocks due to their more limited
capacity to smooth consumption and compensate for shocks. Therefore,
weather shocks might have a distributional effect and exacerbate
existing inequalities magnifying risks of low-birthweight and malnutri­
tion for children born from poorest families that could have long-lasting
consequences (Rosales, 2014).
The negative impact of the shock on yields and consumption might
affect early-life nutrition with potential long-lasting consequences.
Many studies in medical sciences and psychology suggest that low birth
weight and early malnutrition may lead to impaired cognitive devel­
opment (Linnet et al., 2006; Mara, 2003; Shenkin et al., 2004).
Thompson and Nelson (2001) argue that in prenatal months, the
developing brain is vulnerable to external insults including malnutrition
(but also viral infection, alcohol exposure). Rooij et al. (2010) find ev­
idence to this link; exposure to malnutrition in the foetal period during
the Dutch Famine, particularly during the first part of pregnancy,
negatively affects selective attention and inhibitory control later in the
child’s life. Increasingly, the economic literature attempts to quantify
the relationship between health outcomes and subsequent cognitive
achievement (Lo Bue, 2019; Bharadwaj et al., 2018; Sanchez, 2017;
Spears, 2012; Miguel and Kremer, 2004; Glewwe et al., 2001).
The effect of weather shocks on human capital development might
also differ by the child’s sex. A growing body of evidence suggest that
female foetuses are more resilient and adaptive to stress than are male
foetuses (DiPietro and Voegtline, 2017; Rosenfeld, 2015). For instance,
Walsh et al. (2019) find that maternal stress, both physical and psy­
chological distress, increased the probability of pre-term birth, poorer
foetal neurodevelopment (measured by foetal heart rate-movement
coupling), and posed a greater risk to male foetuses. On the other side,
the economic literature has highlighted higher elasticity of human
capital investments in face of rainfall shocks for girls as compared to
boys in India (Chatterjee and Merfeld, 2020; Rose, 1999). Thus, the
unbalanced intra-household allocation of resources in favour of boys
might lead to a more detrimental effect of the shock on girls than on
boys.
In contrast to the effects on cognitive skills, the literature on the
effects of in-utero shocks on non-cognitive skills is almost inexistent. We
are aware of only one study by Krutikova and Lilleor (2015) that ex­
amines the effect of prenatal exposure to rainfall fluctuations on
non-cognitive skills in adulthood in Tanzania, measured using a com­
posite measure of self-esteem, self-efficacy, and locus of control called
core self-evaluation. The authors find that exposure to a 10% increase in
rainfall deviation from the long-run average in-utero increases an in­
dividual’s core self-evaluation by 0.08 standard deviations relative to
their siblings.2
2.3. Biological, nutritional, and behavioural transmission channels
In this section we discuss three transmission channels that might
explain the impact of pre-natal shock on human capital development.
While we cannot distinguish which of these mechanisms may be in play,
we refer to prior literature which may help us theorise the posited
mechanisms. One is through an impact on human brain development
(biological channel) and second is an impact on yields, consumption, and
early-life nutrition (nutritional channel). Furthermore, we recognize the
presence of potential compensating behaviours that might be triggered
by initial shocks (likely mitigating the impact) and potential distribu­
tional effect at play depending on the household wealth and the gender
of the child.
Extensive research in the neuroscience literature has argued that
there are different and potentially critical stages of brain development,
which can have persistent long-term effects on human behaviour.3 Ac­
cording to Stiles and Jernigan (2010), human brain development begins
in the first trimester of pregnancy. From the third to eighth gestational
week (first trimester), rudimentary structures of the brain and central
nervous system are established to form the first well-defined neural
structure.4 The period between the eighth gestational week extending to
approximately mid-gestation is a crucial period in the development of
the neocortex and extends until mid-gestation. The neocortex is
important in higher functions such as sensory perception, spatial
reasoning, conscious thought, and language. In the last trimester of
pregnancy, myelination (fatty insulation of neurons) and synaptogenesis
(forming of synapses between neurons in the nervous system) begin.
According to Thompson and Nelson (2001), all these processes in the
prenatal period are essential to the functional architecture of the brain.
Furthermore, maternal conditions such as stress during the pre-natal
period can play an important role in foetal programming. Psychiatric
studies such as Austin et al. (2005) found that there is an association
between self-reports of maternal trait anxiety, perceived stress, and
depression, and more problematic infant temperament. More recent
research by Aizer et al. (2016) measured stress through the stress hor­
mone cortisol level. The authors show that in-utero exposure to elevated
levels of the stress hormone cortisol has detrimental effects on children’s
educational achievement, cognition, and health.
On the nutritional channels, unexpected variation in rainfall can be
argued to affect early-life conditions through a negative impact on yields
3. Data
This section first, describes the two main source of data used for the
empirical analysis (i.e., the YL data and the rainfall data); second, it
defines the main variables used for this analysis.
3.1. Young Lives
The YL survey is a unique longitudinal cohort study following two
cohorts of children in Andhra Pradesh and Telangana. For this study, we
use the younger cohort data for which we have information since the
first years of life. The younger cohort includes circa 2000 children was
born in 2001–2002 when the children were aged between 6 and 18
months.5 The first study wave was followed by four subsequent rounds
2
Notably, the authors defined the in-utero shock variable using yearly
rainfall data. This approach does not allow to control for seasonality as rainfall
deviations are computed on yearly basis.
3
Knudsen (2004) argues that there are two important periods for the brain
and behaviour: ‘sensitive’ and ‘critical’ periods. ‘Sensitive’ periods are limited
periods during brain development where the effect of experiences on the brain
are unusually strong, while ‘critical’ periods are experiences that occur during
the sensitive period but result in irreversible changes to the brain function.
4
Gestational week refers to the number of weeks post conception (from the
mother’s last menstrual cycle).
5
The older cohort consists of circa 1000 children that were born in
1994–1995 and tracked since about age 8.
3
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
in 2006 (age 5), 2009 (age 8), 2013 (age 12) and 2016 (age 15). The
attrition rate between rounds 1 and 5 is 6%, which is relatively low
compared to other longitudinal studies.
The study sites were selected in 2001 using a semi-purposive sam­
pling strategy to oversample poor households. Hence, YL is not a na­
tionally representative survey.6 The old state of Andhra Pradesh, now
comprising both Andhra Pradesh and Telangana state, was divided into
23 administrative districts, each sub-divided into a number of mandals
(also called sentinel sites or clusters), depending on the size of the district.
In total, there were 1125 mandals with generally between 20 and 40
communities (or villages) in a mandal. The sampling design consisted of
two stages. In the first stage, 20 mandals were chosen based on a set of
economic, human development and infrastructure indicators (Young
Lives, 2017). In the second stage, approximately 100 households with a
child born in 2001–02 were randomly selected from each mandal. The
final sample is spread across 7 districts and 3 regions (Srikakulam and
West Godavari in Coastal Andhra; Anantapur and Kadapa in Rayala­
seema; Karimnagar and Mahbubnagar in Telangana; and Hyderabad),
20 clusters and 100 communities, including both rural and urban
communities.
In all rounds, two main questionnaires were administered to capture
various measurements of child development and other household-level
characteristics: a child questionnaire with data on child health and an­
thropometrics (from age 1),7 cognitive achievements and more specif­
ically receptive vocabulary and numeracy (from age 5 and 8
respectively), non-cognitive skills or personality traits (from age 8) and
other individual characteristics; a household questionnaire (from age 1)
including data on caregiver background, livelihood, demographic
characteristic of household members, socio-economic status, and selfreported shocks. Finally, and most importantly for this paper, YL col­
lects GPS coordinates for all the communities where the YL children live.
This information allows us to estimate with precision the YL children’s
exposure to weather shocks, which will be further explained in Section
3.2.
consists of acknowledging item difficulty and enhancing comparability
over time and across ages (Leon and Singh, 2017).
For non-cognitive skills, YL collects self-reported information about
generalised self-esteem, self-efficacy, and agency measured at age 12
and 15. Self-esteem refers to an individuals’ judgement of their own selfvalue or self-worth and it was measured using the Rosenberg self-esteem
scale (Rosenberg, 1965). It has been found to be correlated to consci­
entiousness and inversely related to the personality trait of neuroticism
(Meier et al., 2011). Self-efficacy is measured through the General
self-efficacy scale (Jerusalem and Schwarzer, 1992) and it refers to the
individual’s belief in the own’s capabilities to produce given attain­
ments and to cope with adversity (Schwarzer and Jerusalem, 1995;
Bandura, 1993). Finally, agency is closely linked to self-efficacy and
builds on the concept of locus of control by Rotter (1966). In this case,
the objective is to measure a child’s sense of agency or mastery over
his/her own life. For each non-cognitive measure, children were asked
to indicate their degree of agreement or disagreement with five state­
ments measured on a Likert scale. The full list of statements and corre­
sponding distribution and raw score reported in the Annex.9
In psychology, self-esteem, self-efficacy, locus of control, and
neuroticism measure a latent personality trait known as “core selfevaluations” (CSE), first examined by Judge et al. (1997). Individuals
with high CSE think positively of themselves and are confident about
their own abilities. Conversely, people with low CSE have a negative
appraisal of themselves and lack confidence. CSE has been found to be
positively correlated with job performance (Judge et al., 1998), the
ability to work in a team (Mount et al., 1995), income level and aca­
demic achievement (Judge and Hurst, 2007). Judge et al. (2003)
developed a core self-evaluation scale including 12 items alike the ones
administered in YL.10 Validation tests show that the measures of
self-esteem, self-efficacy, and agency administered in YL have a high
degree of correlation. A principal component analysis confirms that
items from all three scales load to the first factor which has an eigen­
value of 3.62 and explains 85% of the total variation. The CSE scale
constructed has high internal reliability, with a Cronbach’s alpha of
0.81. This is supported by the psychology literature that questions the
independence of these three related concepts and is cautious about
investigating them in isolation (Judge et al., 2002; Block, 1995). Thus,
we use the first factor emerging from the principal component analysis
as a measure of the latent CSE personality traits. The score is stand­
ardised within the sample.
3.1.1. Cognitive and non-cognitive skills measures
There are two main cognitive indicators used in this analysis:
receptive vocabulary and numeracy skills. Receptive vocabulary is
measured using an adapted version of the Peabody Picture Vocabulary
Test, a widely used test, administered between the ages of 5 and 15 years
old (Dunn and Dunn, 1997). Numeracy skills are assessed using math­
ematics tests developed by YL for the purposes of the survey. The
mathematic tests were not designed to be grade-appropriate but incor­
porate questions at widely differing levels of difficulty: at the basic level,
the tests included questions assessing basic number identification and
quantity discrimination; at the intermediate level, questions on calcu­
lation and measurement; and at the advanced level, questions related to
problem-solving embedded in hypothetical contexts that simulate
real-life situations (e.g., tables in newspapers). The cognitive tests were
collected for all children, regardless of whether they were attending
school or not. This feature of the data avoids the selection problem
which commonly arises when using school-based data.8
The PPVT and the math tests are constructed using Item Response
Theory (IRT) models that are commonly used in international assess­
ments such as PISA and TIMSS. The main advantage of IRT models
3.2. Rainfall data
Rainfall data is obtained from the University of Delaware, possibly
the most commonly use climate dataset in the literature (e.g. used in
Shah and Steinberg, 2017, Rocha and Soares, 2015, and Thai and
Falaris, 2014). It provides gridded climate data on monthly rainfall
precipitations between 1900 and 2014 (Matsuura and Willmott,
2015).11 This long series of data points are used to compute the monthly
historical mean in each of the 100 YL communities in India. To do so, we
match the grid points for which rainfall data was available to the GPS
locations of the YL communities. For each YL community, the survey
collected GPS coordinates using as a reference point the centre of the
community either identified as the centre of the main square or, in
6
Nevertheless, it is shown that the YL sample covers the diversity of children
in poor households in Andhra Pradesh (Kumra, 2008).
7
Considering that only 42.9% of the full sample had their birth weight
recorded, we did not use this variable in the analysis. The sample of children
whose birth weight is reported is quite selected: children whose birth weights
are recorded are socio-economically better off; their mothers have higher ed­
ucation; they live in urban areas; and have fewer siblings.
8
A validation of the psychometric properties of the PPVT and math scores
can be found in Cueto and Leon (2012) and Cueto et al. (2009).
9
The internal consistency of these scales is documented and discussed in
Yorke and Ogando Portela (2018) and Dercon and Krishnan (2009).
10
The full list of items in reported in the Annex, in Table A1.
11
The data can be accessed at the following links at the University of Dela­
ware’s website: Terrestrial Air Temperature: 1900–2014 Gridded Monthly Time
Series (1900–2014) (V 4.01 added 5/1/15) and Terrestrial Precipitation:
1900–2014 Gridded Monthly Time Series (1900–2014) (V 4.01 added 5/1/15).
Each of the values is a local point estimate at a 0.5-degree of longitude-latitude
resolution.
4
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
absence of it, of another point of interests (e.g., city hall, school, post
office, church). The distance from each community GPS location to all
grid points was calculated and the four grid points closest to each YL
community were considered. A distance weight wg was generated for
each grid point g, as follows:
wg =
The defined gestation period accounts for premature births. Information
about premature births and the number of weeks the child was prema­
ture are available in the first survey round as reported by the mother.15
About 9% of mothers (164 observations) reported that their child was
between 1 and 9 weeks premature, with an average of 2 weeks of pre­
maturity. The trimesters of pregnancy were then defined as the periods
between week 0 – 12 (first trimester); week 13—27 (second trimester);
and week 28 until birth (third trimester).
To identify the community of residence of the mother while she was
pregnant with the YL child, we use round 1 information on the com­
munity of residence and information about how long the mother has
been living in the same community. In the attempt to exclude mothers
who may have migrated to the round 1 community of residence to give
birth or after the birth of the child, we exclude from the sample mothers
who reported to have moved to the community while pregnant or after
giving birth, about 6.6% of the sample. Thus, the final sample includes
mothers who lived in the community for at least 2 years and up to 40
years before the first round of data collection, with an average of 9.7
years.16
Information related to the conception date and place of residence
were matched with the relevant rainfall data for the specific community,
month, and year. Therefore, for each child, we defined nine variables,
one for each month m of the gestation period, capturing the monthly
y,m
rainfall deviations RDi,c for the child i, whose mother was living in the
dist−g 1
4
∑
dist−g 1
g=1
with wg ranging from 0 to 1, with grid points closer to the community
having larger weights. The distance weights for the four grid points in
each community summed to 1. For each YL community, the monthly
rainfall precipitation was calculated as a distance-weighted average of
the monthly rainfall registered at the four closest grid points to that
community.
To identify shocks and their severity, we compute the monthcommunity Standardized Precipitation Index (SPI), following Lloyd-­
Hughes and Saunders (2002) methodology. The SPI was first proposed
by McKee et al. (1993) to monitor the severity of droughts in Colorado,
USA.12 The primary advantage of using the SPI is simplicity, since the
rainfall data is the only information needed (i.e., no information about
altitude or soil characteristics are needed). Also, while precipitation is
typically not normally distributed, the SPI normalises the data, making
wetter and drier climates equally represented. Lloyd-Hughes and
Saunders (2002) define rainfall shocks as rainfall fluctuations of at least
1.5 standard deviations away from the historical monthly-andcommunity specific rainfall mean.
Notably, YL collects self-reported data about shocks, information
collected in the first round.13 The survey questions refer to any shock
that occurred since the mother was pregnant, therefore including both
the in-utero period and the first year of life (YL children are on average
11.5 months old in round 1). Despite of that, and the broader definition
of shock used in the self-reported survey module which does not
encompass droughts or floods specifically, we found a substantial cor­
respondence between what the YL households report and the occurrence
of shocks as defined using the rainfall data. More specifically, we found
an overlap between clusters with a high prevalence of households
reporting having been affected by a shock and those hit by strong rainfall
fluctuations during the same period as per the rainfall data. This strongly
suggests that the rainfall shocks registered in the climate data were
indeed perceived and affected the population living in the geographical
area where the shock occurred. However, reliance on external data, as
opposed to self-reported data on shocks, is preferable, as it addresses
concerns of systematic reporting bias besides increasing the precision of
estimates (Cameron and Shah, 2015).
y,m
community c during the years y (either 2000 or 2001). RDi,c is the
difference between the monthly rainfall in the community of residence
and the historical monthly rainfall in the same community:
y,m
m
RDy,m
i,c = Rc − HRc
More specifically, Ry,m
is the rainfall in month m and year y in the
c
community of residence c and HRm
c is the historical rainfall for month m
in the same community c. The historical monthly rainfall is the average
monthly rainfall registered in each community during the period
1900–2014. For instance, the historical average rainfall for January in a
specific community would be computed by averaging out the monthly
rainfall registered in the same community during all the 115 Januaries
during the 1900–2014 period.
Following Lloyd-Hughes and Saunders (2002), we defined as a shock
any monthly rainfall deviation of at least 1.5 standard deviations above
(positive shock or floods) or below (negative shock or droughts) the his­
torical monthly average for the same community.17 To characterize the
intensity of the shock we distinguish between mild shocks (between 1.5
and 2 standard deviation above the historical monthly average) and
strong shocks (any shocks of at least 2 standard deviations above the
historical monthly average). It is worth noticing that computing monthand community-specific rainfall deviation accounts for seasonality, be­
sides identifying communities that are historically more prone than
others to floods and/or droughts.
There is no consensus in the literature on how long the historical
rainfall series (HRm
c ) should be to identify abnormal monthly rainfall.
The length of the historical period varies significantly in the literature,
with averages calculated from as little as 10 years of data (Krutikova and
Lilleør, 2015; Adhvaryu et al. (2018), to the entire range available, often
over 100 years with gridded terrestrial datasets (Dinkelman, 2017;
Webb, 2019; Carrillo, 2020; Yamashita and Trinh, 2020; Baez et al.,
2017 in Mozambique; Andalon et al., 2016 in Colombia). In studies of
3.3. Defining the in-utero period and rainfall shocks
The YL children were born between January 2001 and June 2002.
The date of conception and the gestation period of each YL child is
defined using information about the date of birth and assuming 38
weeks (266 days) as an approximation of a normal-term pregnancy, as
per the World Health Organization definition.14 Therefore, the gesta­
tional period for YL children is between April 2000 and September 2001.
12
SPI is also used by the Indian Meteorological Department for monitoring
purposes (http://www.imdpune.gov.in/hydrology/hydrg_index.html).
13
In 2001/2002 the household head is asked to report any “big changes or
events” that decreased the economic welfare of the household since the mother
was pregnant with the YL child.
14
The World Health Organization define as preterm as giving birth before 37
weeks of pregnancy is completed. See the WHO website: https://www.who.int/
news-room/fact-sheets/detail/preterm-birth. Also, most of the papers use 266
days or 38/40 weeks as threshold to define pre-term pregnancies.
15
There are only 12 cases where the number of weeks the child was premature
is not reported. These observations were deleted from the sample.
16
We cannot exclude that some of the mothers might have spent part of the
pregnancy in a different community as the survey question asks “how long have
you lived in this community for?”, which does not account for temporary shortterm migration.
17
This terminology is used without any specific reference to the intensity of
the rainfall deviation.
5
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
decreasing pattern in the average yearly rainfall during the long his­
torical period suggestive of climate changes significantly affecting
rainfall and seasonality (see Fig. A1 in the Annex). In support to our
argument, Krishnan et al. (2020) report that the annual rainfall series
over Indian landmass shows a decreasing trend, but those trends are not
statistically significant and are irrelevant in the two states studies in our
paper.18
In relation to the weather stations coverage, while there is a clear
worldwide increase in weather station coverage in the first half of the
19th century, the same is follow by a similarly clear decrease after­
wards.19 Furthermore, station coverage varies significantly across
countries and even within the same country. Unfortunately, the weather
data used in this paper do not contain information about the number of
stations contributing to each grid point. The Meteorological Department
of India-IMD (Minister of Earth Science) uses the historical period
1900–2020 to define abnormalities in rainfall (and temperatures). As
there are no clear steering and no evidence suggesting the best historical
period to identify abnormal monthly rainfall, we decide avoiding any
arbitrary cut point and use monthly rainfall data for the entire
1900–2014 period. Nevertheless, we did some robustness checks that
are further discussed in the Results section.
Fig. 1. Average monthly rainfall, historical mean and proportion of children hit
by a rainfall shock during the in-utero period Note: The monthly rainfall re­
ported in the figure (solid line) is computed averaging the monthly rainfall
across the Young Lives communities for the period April 2000-September 2001.
The historical monthly rainfall (dashed line) is the average monthly rainfall
registered in each community during the period 1900–2014. Rainfall is
measured in millimetres and reported on the left y-axis. The bars represent the
proportion of children exposed to a shock (of at least 1.5 SD) in each month
(right y-axis).
4. Rainfall shocks in India during the YL children in-utero
period
According to the Indian Meteorological Department, the climate of
Andhra Pradesh is generally hot and humid. The summer extends from
March to June where moisture level is quite high, the monsoon rainy
season is between June and September (with the pre-monsoon season
between March and May and the post monsoon season between October
and December) and the winter is between October and February
(Guhathakurta et al., 2020).
The dashed line in Fig. 1 displays the average monthly-andcommunity specific historical rainfall registered across the 100 YL
communities and the average monthly-and-community specific rainfall
from April 2000 up until September 2001, corresponding to the in-utero
period of the YL children. The average rainfall is reported in millimetres
and measured on the left y-axis. The bars show the percentage of YL
children affected by a shock of at least 1.5 SD during each in-utero
month, as reported on the right y-axis.
The historical rainfall fluctuations reflect the annual seasonal trend
described above, with the wettest months between June and September,
and the driest months between December and February. Furthermore,
when comparing rainfall fluctuations during the in-utero period against
the historical rainfall we observe that most of the shocks occurred during
summer and the monsoon season in both years but with particularly
intense precipitations happening between March and July 2001. This is
likely due to the cyclone that hit Andhra Pradesh in the monsoon season
in 2000 (De et al., 2005) and extensive flooding reported in the same
areas during the summer 2001 (Ahmed et al., 2013).
The YL communities are distributed across three agro-climatic re­
gions: 42 communities are in Coastal Andhra, 33 in Telangana and 25 in
Rayalaseema. 77% of the children were in rural areas in Round 1%, and
23% in urban areas. Three out of four YL children have been exposed to
Table 1
Prevalence of rainfall shocks of different intensity and nature during pregnancy.
Levels of exposure to rainfall shock
%
Obs.
Affected by a rainfall shock of at least 1.5 SD
Affected by strong rainfall shock (2 SD and above)
Affected by mild rainfall shock (between 1.5 SD and 2 SD)
None
Type of shock (of at least 1.5 SD)
Negative shock only
Positive shock only
Both negative and positive shocks
Affected by a rainfall shock of at least 1.5 SD
First trimester
Second trimester
Third trimester
None
Observations
75.8
30.9
44.9
24.2
1313
535
778
419
12.0
40.8
23.0
75.8
37.0
35.9
31.5
24.2
209
706
398
1313
641
621
546
419
1732
Note: The sample includes all children tracked since round 1 and across the 5
rounds. The sample is constrained to children whose background characteristics
are observed, and at least one of their skills score (PPVT, mathematics or CSE) is
measured in all rounds. Percentage affected by the rainfall shock in-utero by
trimester can overlap since the in-utero shocks can occur in more than one
trimester, depending on the date of conception.
India, the choice seems arbitrary and no explanation is provided in none
of the paper reviewed (Shah and Steinberg, 2017; Kumar et al., 2014).
Reasonably, the choice of the time span used to represent ’normal’
rainfall is specific to the sample used and should take into account for:
first, the potential climate change that might have affected the region
over time and; second, potential poor weather station coverage in early
periods or large changes in stations densities that might lead to artefacts
affecting the trends.
In relation to climate change, we argue that in case of the state of
Andhra Pradesh and Telangana there is not a clear increasing or
18
Decreasing trends in annual rainfall over the period 1901–2015 interests
the regions of Kerala, Western Ghats and some parts of central India, including
Uttar Pradesh, Madhya Pradesh, and Chhattisgarh as well as some parts of the
North-eastern states.
19
The University of Delaware published a figure depicting the number of
(worldwide) weather stations across the whole period (1900–2014), http://
climate.geog.udel.edu/~climate/html_pages/Global2017/PrecipStatNum.pdf).
6
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
Table 2
Comparing children affected and not affected by in-utero shocks.
Child characteristics
Female
Male
Child’s age in months (2016,
Round 5)
Castes
Scheduled Caste
Scheduled Tribe
Backward Caste
Other Caste
Round 1 location (2002)
Rural
Urban
Parent characteristics
Mother’s education
Incomplete primary or less
Completed primary or completed
secondary
Tertiary education and above
Father’s education
Incomplete primary or less
Completed primary or completed
secondary
Tertiary education and above
Exposed to inutero shock
No shocks
Mean
SD
Mean
SD
pvalue
0.47
0.53
179.96
0.01
0.01
0.10
0.45
0.55
180.12
0.02
0.02
0.19
0.554
0.554
0.468
0.18
0.14
0.47
0.21
0.01
0.01
0.01
0.01
0.19
0.20
0.44
0.17
0.02
0.02
0.02
0.02
0.633
0.004
0.341
0.059
0.74
0.26
0.01
0.01
0.87
0.13
0.02
0.02
0.000
0.000
0.71
0.27
0.01
0.01
0.74
0.24
0.02
0.02
0.252
0.202
0.02
0.00
0.02
0.01
0.757
0.56
0.38
0.01
0.01
0.60
0.35
0.02
0.02
0.150
0.254
0.06
1313
0.01
0.05
419
0.01
0.502
Table 4
The effect of in-utero rainfall shocks by intensity, on children’s skills at age 5, 8,
12, and 15.
t-test
Age 5
2 SD and above
F-test, mild = strong shock
Age 8
≥ 1.5 SD and < 2 SD
2 SD and above
F-test, mild = strong shock
Age 12
≥ 1.5 SD and < 2 SD
2 SD and above
F-test, mild = strong shock
Age 15
≥ 1.5 SD and < 2 SD
2 SD and above
F-test, mild = strong shock
Observations
Age 8
Age 12
Age 15
Observations
Math IRT
CSE
-0.154 * *
(0.072)
-0.047
(0.074)
-0.075
(0.089)
-0.126
(0.119)
1313
0.798
(5.935)
-3.492
(5.611)
-13.652 * *
(6.125)
1697
0.080
(0.074)
-0.161 * *
(0.079)
1315
Math IRT
CSE
-0.142 *
(0.081)
-0.176 * *
(0.087)
0.144
-0.056
(0.092)
-0.031
(0.096)
0.041
-0.134
(0.085)
0.029
(0.128)
2.303
-0.220 *
(0.132)
0.041
(0.138)
4.158 * *
1313
-0.861
(6.686)
3.205
(6.661)
0.416
-6.835
(6.042)
1.369
(7.292)
1.290
-13.455 *
(7.198)
-13.938 *
(7.131)
0.004
1697
0.018
(0.092)
0.172
(0.106)
1.470
-0.179 * *
(0.087)
-0.134
(0.100)
0.210
1315
an abnormal amount of precipitation (for at least a month) during the
gestation period (Table 1).20 Furthermore, almost a third of the children
(31%) have been exposed to extreme rainfall shocks during the gestation
period (2 SD or more). The majority of the children who were affected
by a shock experienced positive shock (41% experienced only a positive
shock, 12% experienced only a negative shock, and 23% experienced
both a positive and a negative shock).
Looking at the geographical distribution of the rainfall shocks during
the in-utero period, we find that the prevalence of rainfall shocks (flood
and droughts) during the in-utero period is highest in Telangana com­
munities, while Coastal Andhra and Rayalaseema communities are more
prone to positive shocks compared to the communities in the other two
regions.
In terms of timing, there are some variations in the incidence of
shocks throughout the pregnancy trimesters, but they are roughly
equally spread, with a slightly higher prevalence of rainfall shocks
during the first and second trimesters.
Table 2 reports some basic characteristics comparing children
exposed to a rainfall shock during the gestational period to their peers.
All variables are time-invariant except for the rural/urban location of
residence measured in round 1. By construction, the place of residence
refers to the place where the child was conceived and lived (at least) his/
her first year of life. The p-values for a t-test for differences in means
between the two groups are reported in the last column. Overall, we find
that the exposure to in-utero shocks is homogeneously distributed across
the selected subgroups. While the majority of the YL sample live in rural
areas (77%), we find that children in urban areas are more likely to have
been exposed to shocks in-utero. Although these differences may suggest
that children from more advantaged backgrounds are more likely to be
Table 3
The effect of in-utero rainfall shocks on children’s skills at age 5, 8, 12, and 15.
PPVT IRT
PPVT IRT
Note: All specifications control for child’s age in the specified round, gender, and
caste, year-of-birth fixed-effects, month-of-birth fixed-effects, and community
fixed-effects. Standard errors are clustered at the community level. The sample is
constrained to children of whom their skills scores are observed in every age.
The F-test reports the F-statistic when testing the joint significance that the
coefficient for mild shocks (≥1.5 SD and <2 SD) is equal to strong shocks (2 SD
and above). P-values to show if the estimate is statistically significant from zero
(including the p-values for the F-test) is indicated by * p < 0.1, * * p < 0.05,
* ** p < 0.01.
Note: The sample is constrained to children whose background characteristics
are observed, and at least one of their skills score (PPVT, mathematics or CSE) is
measured in all rounds. Being exposed to rainfall shocks is defined as being
exposed to a rainfall fluctuation of at least 1.5 standard deviations away from the
monthly-community specific historical mean for at least one month during the
gestation period. There is no statistically significant difference in prevalence of
premature births between children who were exposed to the shock in-utero
compared to children who were not exposed (9% and 10% respectively).
Parental education is defined as follows: no education up to grade 6 are
“incomplete primary or less”, grade 7–12 as “completed primary or secondary
education”, and above grade 12 are categorised as “tertiary education and
above”. Scheduled Castes (SCs) and Scheduled Tribes (STs) are traditionally
disadvantaged communities. SCs are the lowest in the traditional caste structure
(formerly known as the ‘untouchables’ and now call themselves Dalit). At the
other end of the spectrum, the Backward Classes (BCs) and the ‘Other Castes’,
the latest also called ‘Upper Castes’ and comprise mostly of ‘forward castes’ who
traditionally enjoy a more privileged socioeconomic status. The p-values for a ttest for differences in means between control group and the treated groups are
reported in the third column.
Age 5
≥ 1.5 SD and < 2 SD
Note: All specifications control for child’s age, gender, and caste, year-of-birth
fixed-effects, month-of-birth fixed effects, and community fixed-effects. Stan­
dard errors are clustered at the community level. The sample is constrained to
children of whom their skills scores are observed at all ages. P-values to show if
the estimate is statistically significant from zero is indicated by * p < 0.1,
* * p < 0.05, * ** p < 0.01.
20
Nearly half of our sample (45%) only experienced the shock for one month
only during the gestation period, 17% experienced for 2 months, 12% for
3 + months.
7
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
affected by shocks, it is reassuring to find that parental education levels
are similar across both groups.
Table 5
Effect of shocks in-utero on skills development, by trimester in which the first inutero rainfall shock occurred.
PPVT IRT
Age 5
1st trimester
2nd trimester
3rd trimester
F-test, trim 1 = trim 2 = trim 3
Age 8
1st trimester
2nd trimester
3rd trimester
F-test, trim 1 = trim 2 = trim 3
Age 12
1st trimester
2nd trimester
3rd trimester
F-test, trim 1 = trim 2 = trim 3
Age 15
1st trimester
2nd trimester
3rd trimester
F-test, trim 1 = trim 2 = trim 3
Observations
-0.133
(0.131)
-0.151
(0.124)
-0.147
(0.125)
0.045
-0.149
(0.140)
-0.131
(0.124)
-0.094
(0.132)
0.145
-0.133
(0.146)
-0.060
(0.132)
-0.199
(0.126)
0.965
-0.273
(0.199)
-0.032
(0.185)
-0.209
(0.167)
1.751
1313
Math IRT
11.174
(9.404)
11.570
(9.095)
3.234
(9.404)
0.835
-12.705
(11.441)
-10.879
(8.720)
-13.741
(9.540)
0.063
-17.812
(12.150)
-17.576 *
(9.767)
-22.571 * *
(9.891)
0.270
1697
5. Empirical approach
CSE
We exploit variations in rainfall across geographic areas (commu­
nity), months and years of birth to identify the causal effect of shock inutero on cognitive and non-cognitive skills development. As mentioned,
the YL children were born in 100 different communities and although
the sampling design was done to identify children of approximately the
same age in round 1. The date of births are spread across 18 months,
between January 2001 and June 2002 as described above.
We test five main hypotheses. First, exposure to rainfall shocks
during the gestational period has long-term effects on cognitive and noncognitive skills development throughout childhood and adolescence.
Second, the effect of in-utero exposure to rainfall shocks increases with
the intensity of the shock. Third, whether the effect of the shock in­
creases monotonically with the number of shocks that occurred during
the gestation period. Four, the effect of the in-utero exposure to rainfall
shock is time-sensitive, i.e. it depends on which trimester of pregnancy
the shock occurred. Five, in-utero shocks have a distributional effect by
gender and socio-economic status.
The effect of in-utero rainfall shocks on children’s future outcomes is
specified as follows:
0.032
(0.148)
-0.042
(0.129)
-0.016
(0.119)
0.332
-0.125
(0.152)
-0.087
(0.143)
-0.086
(0.127)
0.138
1315
Yijc,t =
∝+
β0 Sijc + γCj + ωc + yobi + mobi + εij,t
(1)
where Yijc,t is the outcome of interest of child i, at age t, born in the
household j and whose mother was living in community c during preg­
nancy. The outcomes measured are PPVT scores at ages 5, 8, 12, and 15;
mathematics scores at ages 8, 12, and 15; and CSE scores at ages 12 and
15. Sijc is the shock variable and it takes a value equal to 1 if for at least
one month during the gestation period the community where the mother
of the child was living was exposed to a rainfall shock of at least 1.5
standard deviations, as calculated by the SPI. The main parameter of
interest is β0 which captures the impact of in-utero shocks on the child’s
outcome. As long as the rainfall shock is exogenous, that is E(Sijc , εij,t )
= 0, β0 is unbiased and provides the causal effect of a rainfall shock on
Yijc,t . This will be discussed further at the end of this section.
The vector C includes the child’s age in months, gender, and his/her
caste. This specification also includes maternal community of residence
fixed-effects ωc that are intended to control for any unobservable (timeinvariant) community-specific characteristics, that might make some
communities more prone to weather shocks or more disadvantaged than
others in term of health and education inputs (such as the availability
and quality of health services, prenatal care, and education services).
The ideal geographical level to be used for the fixed-effect is the com­
munity, given that the rainfall variable is defined at the community
level.21 Finally, we include a year of birth fixed-effect, yobi , and month
of birth fixed-effect, mobi , to account for time trends and potential
cofounding effects of being born in a certain month, as discussed below.
εij,t is an idiosyncratic error term. In the regressions, standard errors are
clustered at the community level.
To capture the heterogeneity of the effect of shocks in-utero we
investigate how it varies depending on: first, its intensity (i.e., whether
the rainfall shock is of at least 1.5 standard deviations or 2 + standard
deviations); second, its duration or frequency (i.e., the number of
monthly shocks occurred in-utero); and third, its timing (i.e., during
which pregnancy trimester the first shock occurs).
In Eq. (2), Ik corresponds to three k levels of intensity of the rainfall
Note: 1st trimester is the gestation period between week 0—12, 2nd trimester is
week 13 – 27, 3rd trimester is between week 28 – birth. All specifications control
for child’s age in the specified round, gender, and caste, total number of shocks
that occurred in-utero, year-of-birth fixed-effects, month-of-birth fixed effects,
and community fixed-effects. Standard errors are clustered at the community
level. The sample is constrained to children of whom their skills scores are
observed in every age. The F-test reports the F-statistic when testing the joint
significance that the coefficient for each of the trimesters are equal. P-values to
show if the estimate is statistically significant from zero (including the p-values
for the F-test) is indicated by * p < 0.1, * * p < 0.05, * ** p < 0.01.
Fig. 2. Estimates on exposure 0–7 years before birth on PPVT, Mathematics and
CSE scores at age 15 Note: IU refers to the 9-months in-utero period. Figure plots
the estimated coefficients on any rainfall shock occurred in the 9-months period
in-utero and pre-conception (up to 7 years) on children’ skills at round 5. The
estimated coefficients correspond to separated specifications on each of the
three scores at age 15 for each 9-monthsperiod before birth, with the same
controls specified in the baseline regression in Table 3, and clustered at the
community level. The squares represent p-value at the 5% significance level,
and the diamonds represent p-values at the 1% significance level.
21
The relatively small sample size might raise concerns about the limited
within-community variation with the fixed-effect capturing most of the varia­
tion in the data. However, results are similar when fixed-effect at YL cluster
level are considered.
8
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
shock (0 = no shock, the base category; 1 = mild shock; 2 = strong
shock). Thus, the parameters of interest, σ 1 and σ 2 , correspond to the
effect of being exposed to a mild shock and a strong shock, respectively,
compared to children who did not experience any shocks in utero.
2
∑
Yijc,t =
∝+
σk I kijc + γCj + ωc + yobi + mobi + εij,t
6. Results
Table 3 shows the average effects of experiencing a rainfall shock at
any point during the in-utero period on children’s cognitive and noncognitive skills. Like previous studies (Almond et al., 2015; Almond
and Mazumder, 2011), we find that being exposed to a rainfall shock
reduces children’s cognitive skills. In particular, we find that being
exposed to a rainfall shock in-utero reduces PPVT scores at age 5 (in 0.15
points or 5% lower score respect to the control group) and the math
scores (in 13.6 points or 2% lower score respect to the control group) at
age 15. Additionally, we find novel evidence that rainfall shocks in-utero
reduces children’s non-cognitive skills at age 15 by 0.16 points. These
results are robust to alternative definition of the historical period length
used to compute a “normal” monthly rainfall.23
Besides PPVT, the effects of an in-utero shock seem to only manifest
at later ages. While we cannot empirically prove it and fully explain why
this occurs, Almond and Currie (2011) argue that there may be “latent
disadvantages” which are unobserved, i.e. events in-utero can alter the
infant in a way that leads to later disease even in infants who are
apparently healthy at birth. The most recent study by Conti et al. (2020)
suggests that events occurring in the prenatal period can influence
postnatal development without affecting early measured human capital
(in most of the cases measured by birth weight). Similarly, we argue that
the stronger effects on skills observed at age 15 may reflect the existence
of “latent disadvantages” that manifest (or are measurable) only at older
ages. Latent disadvantage might not affect the child’s ability to compute
basic arithmetic at young ages, but possibly more complex and higher
abilities measured during adolescence.
The negative impact on children’s PPVT at age 5 and mathematics at
age 15 are driven by boys (see Table A4 in the Annex). Boys who
experienced a 1.5 standard deviation shock in-utero had 0.3 standard
deviations lower PPVT, significant at the 5% level, whereas the effect on
girls are close to zero. Boys who experienced a shock had lower math­
ematics scores at age 15, significant at the 5% level. Girls experienced
lower mathematics scores too, but the effect is not statistically signifi­
cant. Finally, we did not find any gender differential effects on noncognitive scores.
Furthermore, we find evidence that in-utero shocks affect both
cognitive and non-cognitive skills of children of lower educated parents
relatively more than children whose parent have at least completed
primary education (see Table A5 in the Annex). The effect of a shock
reduces mathematics and CSE scores at age 15, significant at the 5%
level for lower educated parents. Hence, weather shocks might have a
distributional effect exacerbating pre-existing inequalities.
We then investigate the effect of being exposed to a rainfall shock
during the pre-natal period varies according to the intensity of the
shock. More specifically, Table 4 shows the effect of a mild shock and of
a strong shock compared to children who have not been exposed to a
shock in-utero. As reported in Table 1, strong shocks (2 SD or more) are
relatively less common that mild shocks as (31% compared to 43% of the
sampled have been exposed respectively to strong and mild rainfall
shocks during the gestation period). According to our results, both mild
and strong shocks have an equally negative effects on PPVT at age 5 and
mathematics scores at age 15, the two estimated parameters being not
statistically different to one another (Table 4). Furthermore, the nega­
tive impacts on PPVT and CSE scores at age 15 seem to be driven by
milder shocks, as stronger shocks do not have an impact on those skills at
the same age.
(2)
k=0
In Eq. (3), we examine whether the effect of the shock is monotonic
according to the duration or frequency of shocks in-utero. We include an
indicator variable Dl where l represents different categories of shocks
durations 0 = no shock, the base category; and the number of shocks
that happened during the in-utero period which are categorised as 1, 2,
and 3 or more, since 45% of the sample experienced one shock only,
17% experienced 2 shocks and 2% experienced 3 or more shocks.
3
∑
Yijc,t =
∝+
πk Dlijc + γCj + ωc + yobi + mobi + εij,t
(3)
l=0
In Eq. (4), we explore whether there are key periods during preg­
nancy when exposure to rainfall shocks are more likely to affect the
child’s skills development. Given that in our sample some children have
been affected by shocks in more than one trimester, we analyse the effect
of the shock timing by identifying the effect of the trimester (first, sec­
ond or third) when a shock first occurred. To do so, we include the
variables Sr for whether the shock happened in trimester one, two or
three or never happened (base category). We also control for the total
number of monthly shocks that occurred throughout pregnancy, indi­
cated by Tijc .
4
∑
Yijc,t =
∝+
φr Srijc + δTijc + γCj + ωc + yobi + mobi + εij,t
(4)
r=0
Finally, we investigate the distributional effect of an in-utero shock
by the child’s gender and the household socio-economic status. To do so,
we estimate Eq. (1) separately for boys and girls and distinguishing
households where the highest educational level completed by the par­
ents corresponds to having completed (or not) primary education.
To identify the effects of in-utero rainfall shocks on children’s skills,
we rely on the assumption that rainfall shocks are random, which seems
to be the case, as shown in Table 2. Furthermore, an underlined
assumption is that the decision to get pregnant is not timed according to
seasonality. As reported in Fig. A2 in the Annex, the date of births are
spread across 18 months, between January 2001 and June 2002 and
tend to be slightly more common during the monsoon season.22 The
inclusion of month-of-birth fixed effect would control for potential
cofounding effect. Furthermore, if mothers did time their pregnancies,
then we would find differences in the background characteristics of
mothers who gave birth in the monsoon period compared to those who
gave birth in a different period, which would invalidate our assumption
that rainfall shocks and pregnancy are orthogonal. Table A2 in the Ap­
pendix reports the average background characteristics of mothers of
children born in the monsoon compared to those not born in the
monsoon period. It is reassuring to find that the characteristics of the
mothers (education; health, proxied by height; and, location and the
number of children) in the two subgroups are largely similar, the only
exception being mother’s age, but the difference is small by about half a
year. A final concern would be if mothers were able to time the preg­
nancy anticipating weather shocks. However, it seems quite unlikely, as
it would require sophisticated forecasts models.
23
Guttman (1999) recommends using a period of at least 50-year to compute
SPI values, for computational accuracy. We perform some robustness check
using an historical period between 115- and 50 years. Overall, our results seem
robust when using different data spans of 50 years+ , though less precise as
reported in Table A3 in the Annex for an historical period of 50 years of
monthly rainfall data.
22
Notably, the gestation period for virtually all children in the sample overlap
with the monsoon season for at least a month. Only 2 children (0.11% of the
sample) were not exposed to the monsoon period at any month during the inutero period.
9
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
Similarly, as for the intensity of the shocks, we find little evidence of
a non-monotonic relationship between the effect of pre-natal shocks on
skills development. According to our results, being exposed to one, two,
or three or more shocks does not have a differential impact on skills (see
Table A6 in the Annex).
Overall, these estimates suggest that being exposed to in-utero
shocks, but not the intensity or the frequency of the shock, affects
children’s future skills development. Arguably, the sample size of the YL
children may be relatively small which lack power identifying variation
in these effects by shock intensity and frequency.24
Finally, we investigate whether the timing of the shock matter. More
specifically, we explore whether the impacts of being exposed to rainfall
shocks differ depending on whether the child was exposed for the first
time to a rainfall shock in the first, second or third pregnancy trimester.
Table 5 shows that in-utero rainfall shocks during the second or third
trimester negatively affect the math score at age 15, while no significant
effect is found for shocks occurring in the first trimester. Nevertheless,
the joint F-test shows that the coefficient estimates are statistically
indistinguishable from one another. We do not find any differences in
the effect of the shocks by trimester on PPVT scores nor CSE scores.
One concern is that the negative effects of in-utero exposure to
rainfall shocks on skills development may be confounded with omitted
variables. To verify that this is not the case, we estimate the effects of
shocks (of at least 1.5 SD) occurring in each 9-months period before the
child’s conception date up to 7 years (or 63 months), on children’s
mathematics, PPVT, and CSE scores at age 15.
In Fig. 2, each marker corresponds to the estimated parameter
capturing the effect of the shock happening in each 9-months periods inutero (i.e., coefficients reported in Table 3) and before conception on the
child’s skills score. If the results presented in the previous section were
spurious and driven by omitted variables, the results in Fig. 2 and those
in the previous section should be similar. Overall, there is no such evi­
dence.25 Fig. 2 shows evidence of no significant effects of rainfall shocks
on PPVT, Maths, and CSE scores before conception. This suggests that
the effect of rainfall on children skills development is only relevant
during the in-utero period and that there are no other mechanisms (e.g.,
maternal nutrition) through which previous rainfall shocks might affect
children’s skills development.
boys, while the effect for both cognitive and non-cognitive skills are
driven by children of parents with lower education, suggesting that
prenatal shocks might exacerbate pre-existing inequalities.
The most important limitation of this study is the inability to
investigate the mechanisms through which the unveiled effects arise. As
discussed in the paper the effects of rainfall shocks on child’s develop­
ment can be explained by changes in crops production, prices, and in­
come with an impact on consumption and nutrition or by a
disproportionate amount of stress caused to the mother and interfering
directly on brain development in very critical period. Finally, the (net)
impact of the shock will depend on the compensating mechanisms
intervening in mitigating the detrimental effect of the shock with po­
tential distributional effects. Our results suggest that, regardless of any
investment strategy the household might put in place, being exposed to a
weather shock during the in-utero period has a negative long-lasting
effect on skills.
A potential threat for identification in our analysis, as in previous
similar studies, concerns the mortality of weak foetuses due to adverse
weather conditions. In fact, to the extent that rainfall shocks increase
foetal mortality, the population of new-borns included in our sample
would include those who survived to the shock. If this were the case, the
effect of the in-utero-shock would be underestimated (Andalon et al.,
2016). While the magnitude of the potential bias cannot be established,
we argue that our estimates would represent a lower bound of the real
effect of rainfall shock in-utero.
Moreover, the relatively reduced sample of YL might represent a
limitation. For instance, it might be that the estimated statistically
insignificant effects on cognitive and non-cognitive skills are the product
of a lack of statistical power. However, the longitudinal feature of the YL
data and the fact that it includes several measures of children’s skills
represent an important advantage of this dataset in comparison to crosssectional or administrative datasets with larger sample sizes.
Climate change and other negative shocks (e.g., pandemics) are
likely to happen more often in the future. Given the importance of skills
in determining educational, labour, and social outcomes, and the
importance of early skills development, policies should be designed to
protect maternal welfare during pregnancy. This could be through cash
or in-kind (e.g., food) transfers for mitigate the impacts of the shock on
the household wealth but also offering psychological support to mothers
who are dealing with the stress and anxiety caused by the shock during
the delicate phase of pregnancy.
8. Discussion
Acknowledgements
Differences in education, labour, and social outcomes later in life can
be originated at very early stages. In this paper, we analyse the impor­
tance of in-utero conditions for the formation and development of
cognitive and non-cognitive skills in a sample of children and teenagers
in India. More specifically we exploit variations in rainfall across
geographic areas (community), months, and years of birth to identify the
causal effect of shock in-utero on cognitive and non-cognitive skills
development throughout childhood and adolescence. Our results indi­
cate that being exposed to rainfall shocks when in-utero are detrimental
to children’s receptive vocabulary at age 5, and on mathematics and
non-cognitive skills at age 15. The effect for cognitive skills is driven by
This is research has been funded by Foreign, Commonwealth &
Development Office, grant number 200425.
7. Falsification tests
Appendix
See Appendix Fig. A1, Fig. A2.
See Appendix Tables A1–A6.
24
While not shown here, note that in Table 1, most of the shocks were positive
or both positive and negative shocks. We find that our estimates are largely
driven by positive shocks, given the higher prevalence. The estimated param­
eters for positive and negative shocks are not statistically significant from one
another (results not reported, available on request).
25
Results from a multiple hypothesis testing confirm these results. We
calculated the Westfall-Young stepdown adjusted p-values, which control for
the probability of making any Type I error (i.e., a false positive). Results are
available upon request.
10
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
Fig. A1. Inter-annual variability in rainfall during 1900–2014 in the Young Lives communities (Average yearly rainfall 1900–2014).
Fig. A2. Date of birth distribution.
Table A1
Definitions of non-cognitive skill items used.
Agency: Individual’s sense of agency or mastery over his/her own life
1) I have no choice about the work I do
2) If I study hard, I will be rewarded with a better job in the future
3) I like to make plans for my future studies and work
4) Other people in my family make all the decisions about how I spend my time
5) If I try hard, I can improve my situation in life
Self-esteem (Rosenberg Scale): Individuals’ judgement of their own self-value or self-worth
1) I do lots of important things
2) In general, I like being the way I am
3) Overall, I have a lot to be proud of
4) I can do things as well as most people
5) Other people think I am a good person
6) A lot of things about me are good
7) I’m as good as most other people
8) When I do something, I do it well
Generalised Self-efficacy Scale: One’s belief in their capabilities to produce given attainments and to cope with adversity
1) I can always manage to solve difficult problems if I try hard enough.
2) If someone opposes me, I can find the means and ways to get what I want.
3) It is easy for me to stick to my aims and accomplish my goals.
4) I am confident that I could deal efficiently with unexpected events.
5) Thanks to my resourcefulness, I know how to handle unforeseen situations.
6) I can solve most problems if I invest the necessary effort.
7) I can remain calm when facing difficulties because I can rely on my coping abilities.
8) When I am confronted with a problem, I can usually find several solutions.
9) If I am in trouble, I can usually think of a solution.
10) I can usually handle whatever comes my way.
Core Self-evaluation: Core self-evaluation is a trait that reflects an individual’s evaluation of their abilities and own control (Judge et al., 1998). It is predicted from a principal factor
analysis of the agency scale, self-esteem scale and generalised self-efficacy scale, all standardised to mean zero and standard deviation of one.
11
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
Table A2
Maternal and household characteristics of children born or not born in monsoon season.
Born in monsoon
Mother’s education
Incomplete primary or less
Completed primary or up to completed secondary
Tertiary education and above
Urban location
Mother’s height
Mother’s age (years)
Number of older siblings
Number of older sisters
Number of older brothers
Ethnicity
Scheduled Caste
Scheduled Tribe
Backward Caste
Other
Not born in monsoon
T-test
Obs.
Mean
SE
Mean
SE
p-value
0.72
0.25
0.02
0.23
151.38
24.11
0.72
0.51
0.45
0.018
0.017
0.006
0.017
0.232
0.180
0.039
0.038
0.037
0.71
0.27
0.02
0.23
151.62
23.51
0.73
0.52
0.47
0.014
0.013
0.004
0.013
0.185
0.129
0.031
0.030
0.029
0.565
0.456
0.638
0.948
0.439
0.006
0.724
0.864
0.789
1732
1732
1732
1732
1717
1728
1732
1732
1732
0.16
0.48
0.20
0.72
0.015
0.015
0.020
0.016
0.15
0.46
0.20
0.73
0.012
0.011
0.015
0.012
0.715
0.595
0.831
0.724
1732
1732
1732
1732
Note: SE = Standard error. The sample is constrained to children whose background characteristics are observed, and at least one of their skills score (PPVT,
mathematics or CSE) is measured in all rounds. The p-values for a t-test for differences in means between control group and the treated groups are reported in the
second column.
Table A3
The effect of in-utero rainfall shocks on children’s skills (historical period: 1950 − 2002).
Age 5
Age 8
Age 12
Age 15
N
PPVT IRT
Math IRT
CSE
-0.139 *
(0.077)
-0.004
(0.072)
-0.077
(0.092)
-0.078
(0.126)
1313
2.830
(5.809)
-1.043
(5.982)
-9.494
(6.362)
1697
0.027
(0.081)
-0.158 *
(0.081)
1315
Note: Historical period defined using monthly rainfall data for the period 1950–2002. All specifications control for child’s age, gender, and caste, year-of-birth fixedeffects, month-of-birth fixed effects, and community fixed-effects. Standard errors are clustered at the community level. The sample is constrained to children of whom
their skills scores are observed at all ages. P-values to show if the estimate is statistically significant from zero is indicated by * p < 0.1, * * p < 0.05, * ** p < 0.01.
Table A4
In-utero rainfall shocks on children’s skills, by child’s gender.
PPVT IRT
Age 5
Age 8
Age 12
Age 15
N
Math IRT
CSE
Male
Female
Male
Female
Male
Female
-0.326 * *
(0.127)
-0.104
(0.109)
-0.096
(0.132)
-0.242
(0.163)
687
0.004
(0.129)
-0.035
(0.128)
-0.100
(0.139)
-0.160
(0.173)
626
-3.782
(7.832)
-12.316
(7.794)
-15.236 * *
(7.168)
912
4.174
(8.990)
5.664
(8.682)
-14.169
(10.697)
785
0.057
(0.105)
-0.049
(0.112)
711
-0.022
(0.137)
-0.252
(0.163)
604
Note: All specifications control for child’s age in the specified round, gender, and caste, year-of-birth fixed-effects, month-of-birth fixed-effects, and community fixedeffects. Standard errors are clustered at the community level. The sample is constrained to children of whom their skills scores are observed in every age. P-values to
show if the estimate is statistically significant from zero is indicated by *p < 0.1, * *p < 0.05, * **p < 0.01.
12
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
Table A5
In-utero rainfall shocks on children’s skills, by highest parental education.
PPVT IRT
Age 5
Age 8
Age
12
Age
15
N
Math IRT
CSE
Less than completed
primary education
Completed primary
and above
Less than completed
primary education
Completed primary
and above
Less than completed
primary education
Completed primary
and above
-0.140
(0.111)
-0.107
(0.109)
-0.147
-0.125
(0.151)
0.043
(0.114)
0.039
-9.786
(9.725)
-6.644
15.930 *
(8.157)
5.937
0.138
0.009
(0.148)
-0.248
(0.135)
-0.055
(10.846)
-24.825 * *
(8.049)
0.697
(0.120)
-0.239 * *
(0.116)
-0.091
(0.200)
761
(0.196)
552
(11.363)
862
(8.874)
835
(0.119)
653
(0.106)
662
Note: All specifications control for child’s age in the specified round, gender, and caste, year-of-birth fixed-effects, month-of-birth fixed-effects, and community fixedeffects. Standard errors are clustered at the community level. The sample is constrained to children of whom their skills scores are observed in every age. P-values to
show if the estimate is statistically significant from zero is indicated by *p < 0.1, * *p < 0.05, * **p < 0.01.
Table A6
The effect of in-utero rainfall shocks by frequency, on children’s skills at age 5,8,12 and 15.
Age 5
1 shock
2 shocks
≥ 3 shocks
F-test, 1 shock = 2 shocks = ≥3 shocks
F-test, 1 shock = 2 shocks
F-test, 1 shock = ≥3 shocks
F-test, 2 shocks = ≥3 shocks
Age 8
1 shock
2 shocks
≥ 3 shocks
F-test, 1 shock = 2 shocks = ≥3 shocks
F-test, 1 shock = 2 shocks
F-test, 1 shock = ≥3 shocks
F-test, 2 shocks = ≥3 shocks
Age 12
1 shock
2 shocks
≥ 3 shocks
F-test, 1 shock = 2 shocks = ≥3 shocks
F-test, 1 shock = 2 shocks
F-test, 1 shock = ≥3 shocks
F-test, 2 shocks = ≥3 shocks
Age 15
1 shock
2 shocks
≥ 3 shocks
F-test, 1 shock = 2 shocks = ≥3 shocks
F-test, 1 shock = 2 shocks
F-test, 1 shock = ≥3 shocks
F-test, 2 shocks = ≥3 shocks
N
PPVT IRT
Math IRT
CSE
-0.157 * *
(0.072)
-0.117
(0.109)
-0.073
(0.169)
0.134
0.875
0.184
0.669
-0.051
(0.073)
0.033
(0.121)
0.092
(0.154)
0.555
0.662
1.061
0.287
-0.078
(0.089)
-0.018
(0.135)
0.222
(0.158)
3.108 * *
0.410
5.485 * **
4.982 * **
-0.127
(0.119)
-0.106
(0.165)
0.098
(0.257)
0.565
0.028
1.040
1.031
1313
0.354
(5.952)
5.326
(8.904)
-1.729
(10.473)
0.772
0.439
0.050
1.247
-3.965
(5.609)
5.603
(8.958)
17.802
(11.771)
2.028
1.618
3.854 *
1.409
-14.212 * *
(6.149)
-3.987
(9.364)
5.429
(14.766)
1.247
1.783
2.075
0.687
1697
0.079
(0.075)
0.081
(0.114)
0.193
(0.170)
0.325
0.000
0.500
0.629
-0.150 *
(0.081)
-0.337 * **
(0.115)
-0.121
(0.168)
2.367 *
3.630 *
0.039
2.355
1315
Note: All specifications control for child’s age in the specified round, gender, and caste, year-of-birth fixed-effects, month of birth fixed-effects, and community fixedeffects. Standard errors are clustered at the community level. The sample is constrained to children of whom their skills scores are observed in every age. The F-test
reports the F-statistic when testing the joint significance that all coefficients are equal, as well as equality of coefficients between the categorical number of shocks. Pvalues to show if the estimate is statistically significant from zero (including the p-values for the F-test) is indicated by * p < 0.1, * * p < 0.05, * ** p < 0.01.
13
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
References
Dillon, A., McGee, K., Oseni, G., 2015. Agricultural production, dietary diversity and
climate variability. J. Dev. Stud. 51 (8), 976–995.
Dinkelman, T., 2017. Long-run health repercussions of drought shocks: evidence from
south african homelands. Econ. J. 127, 1906–1939. https://doi.org/10.1111/
ecoj.12361.
DiPietro, K., Voegtline, M., 2017. The gestational foundation of sex differences in
development and vulnerability. Neuroscience 3422017, 4–20.
Dunn, L.M., Dunn, L.M., 1997. Peabody picture vocabulary test-III. Circle Pines MN.
American Guidance Service.
Fletcher, J.M., 2018. The effects of in-utero exposure to the 1918 influenza pandemic on
family formation. Econ. Hum. Biol. 30, 59–68. https://doi.org/10.1016/j.
ehb.2018.06.004.
G. Conti M. , Hanson H. , Inskip S. , Crozier C. , Cooper K. , Godfrey Beyond birthweight:
the origins of human capital IZA Work. Pap. 2020.
Glewwe, P., Jacoby, H.G., King, E.M., 2001. Early childhood nutrition and academic
achievement: a longitudinal analysis. J. Public Econ. 81 (3), 345–368. https://doi.
org/10.1016/S0047-2727(00)00118-3.
Guhathakurta, P., Sanap, S., Menon, P., Prasad, A.K., Sangwan, N., Advani S.C. , 2020.
Observed Rainfall Variability and Changes over Andhra Pradesh state. Published by
Government of India. Ministry of Earth Sciences. India Meteorological Department.
Guttman, N.B., 1999. Accepting the Standardized Precipitation Index: a calculation
algorithm. J. Am. Water Resour. Assoc. 35 (2), 311–322.
Heckman, J.J., Stixrud, J., Urzua, S., 2006. The effects of cognitive and noncognitive
abilities on labor market outcomes and social behavior. J. Labor Econ. 24 (3),
412–482. https://doi.org/10.1086/504455.
Hoddinott, J., 2006. Shocks and their consequences across and within households in
rural Zimbabwe. J. Dev. Stud. 42 (2), 301–321. https://doi.org/10.1080/
00220380500405501.
J. Chatterjee J.D. Merfeld Protecting Girls from Droughts with Social Safety Nets IZA
Discuss. 2020.Paper No. 13694. Bonn.
Jerusalem, M., Schwarzer, R., 1992. Self-efficacy as a resource factor in stress appraisal
processes. In: Schwarzer, R. (Ed.), Self-efficacy: Thought Control of Action.
Hemisphere Publishing Corp, pp. 195–213.
Judge, T.A., Hurst, C., 2007. The benefits and possible costs of positive core selfevaluations: a review and agenda for future research. In: Nelson, D., Cooper, C.L.
(Eds.), Positive Organizational Behavior. Sage Publications, London, UK,
pp. 159–174. https://doi.org/10.4135/9781446212752.n12.
Judge, T.A., Locke, E.A., Durham, C.C., 1997. The dispositional causes of job satisfaction:
a core evaluations approach. Res. Organ. Behav. 19, 151–188.
Judge, T.A., Erez, A., Bono, J.E., 1998. The power of being positive: The relation between
positive self-concept and job performance. Hum. Perform. (2–3), 167–187. https://
doi.org/10.1080/08959285.1998.9668030.
Judge, T.A., Erez, A., Bono, J.E., Thoresen, C.J., 2002. Are measures of self-esteem,
neuroticism, locus of control, and generalized self efficacy indicators of a common
core construct? J. Personal. Soc. Psychol. 83, 693–710. https://doi.org/10.1037/
0022-3514.83.3.693.
Judge, T.A., Erez, A., Bono, J.E., Thoresen, C.J., 2003. The Core Self-Evaluations Scale
(CSES): development of a measure. Pers. Psychol. 56, 303–331. https://doi.org/
10.1111/j.1744-6570.2003.tb00152.x.
Knudsen, E.I., 2004. Sensitive Periods in the Development of the Brain and Behavior.
J. Cogn. Sci. 16 (8), 1412–1425. https://doi.org/10.1162/0898929042304796.
Krishnan, R., Sanjay, J., Gnanaseelan, C., Mujumdar, M., Kulkarni, A., Supriyo
Chakraborty, 2020. Assessment of climate change over the Indian region. A Report
of the Ministry of Earth Sciences (MoES). Government of India.
Kumar, S., Molitor, R., Vollmer, S., 2014. Children of Drought: Rainfall Shocks and Early
Child Health in Rural India, No 1407, Working Papers, Sam Houston State
University, Department of Economics and International Business. https://
EconPapers.repec.org/RePEc:shs:wpaper:1407.
Kumra, N. , 2008, “An Assessment of the Young Lives Sampling Approach in Andhra
Pradesh, India”, Young Lives Technical Note No. 2, March 2008.
Leon, J., Singh, A. , 2017. Equating test scores for receptive vocabulary across rounds and
cohorts in Ethiopia, India and Vietnam. Young Lives Technical Note (40).
Linnet, K.M., Wisborg, K., Agerbo, E., Secher, N.J., Thomsen, P.H., Henriksen, T.B., 2006.
Gestational age, birth weight, and the risk of hyperkinetic disorder. Arch. Dis. Child
91 (8), 655–660. https://doi.org/10.1136/adc.2005.088872. PMID: 16754656;
PMCID: PMC2083047.
Lloyd-Hughes, B., Saunders, M.A., 2002. A drought climatology for Europe. Int. J.
Climatol. 22, 1571–1592. https://doi.org/10.1002/joc.846.
Lo Bue, M.C., 2019. Early childhood during indonesia’s wildfires: health outcomes and
long-run schooling achievements. Economic Development and Cultural Change.
University of Chicago Press, pp. 969–1003. https://doi.org/10.1086/700099.
Maccini, S., Yang, D., 2009. Under the weather: health, schooling, and economic
consequences of early-life rainfall. Am. Econ. Rev. 99 (3), 1006–1026. https://doi.
org/10.1257/aer.99.3.1006.
Majid, M.F., 2015. The persistent effects of an in utero nutrition shocks over the life
cycle: evidence from Ramadan fasting. J. Dev. Econ. 117 (Issue C), 48–57. https://
doi.org/10.1016/j.jdeveco.2015.06.006.
Mara, D., 2003. Water, sanitation and hygiene for the health of developing nations.
Public Health 117 (6), 452–456. https://doi.org/10.1016/S0033-3506(03)00143-4.
Matsuura K. , Willmott C.J. , Terrestrial Precipitation: 1900-2014 Gridded Monthly Time
Series Version 4 2015 01.〈http://climate.geog.udel.edu/~climate/html_pages/
Global2014/README.GlobalTsP2014.html〉.
McKee T.B. , Doesken N. , Kleist J. , The relationship of drought frequency and duration
to time scales Prepr., 8th Conf. Appl. Climatol. 1993 179 184.
Adhvaryu, A., Nyshadham, A., Molina, T., Tamayo, J., 2018. Helping children catch up:
early life shocks and the PROGRESA experiment. NBER Work. Pap. Ser. 24848.
Ahmed, S., Ray, R., 2017. Do in utero shocks have adverse effects on child health
outcomes and can welfare schemes ameliorate such effects? Evidence from Andhra
Pradesh, India. J. Biosoc. Sci. 50 (6), 770–799. https://doi.org/10.1017/
S0021932017000591.
Ahmed, Z., Mohan Rao, D.R., Mohan Reddy, K.R., Raj, Y.E., 2013. Urban Flooding: Case
of Hyderabad. Glob. J. Eng. Des. Technol. Vol 2 (4), 63–66.
Aizer, A., Stroud, L., Buka, S., 2016. Maternal stress and child outcomes: evidence from
siblings. J. Hum. Resour. 51 (3), 523–555. https://doi.org/10.3386/w18422.
Almlund, M., Duckworth, A.L., Heckman, J.J., Kautz, T.D., 2011. In: Hanushek, E.,
Machin, S., Woessman, L. (Eds.), Personality and Psychology Economics. Handbook
of the Economics of Education, Amsterdam, pp. 1–181.
Almond, D., 2006. Is the 1918 influenza pandemic over? Long-term effects of in utero
exposure in the post-1940 U.S. Population. J. Political Econ. 114 (4), 672–712.
https://doi.org/10.1086/507154.
Almond, D., Currie, J., 2011. Killing me softly: the fetal origins hypothesis. J. Econ.
Perspect. 25 (3), 153–172. https://doi.org/10.1257/jep.25.3.153.
Almond, D., Mazumder, B., 2011. Health capital and prenatal environment: the effect of
ramadan observance during pregnancy. Am. Econ. J.: Appl. Econ. 3, 56–85. https://
doi.org/10.1257/app.3.4.56.
Almond, D., Mazumder, B., Ewijk, R., 2015. In utero Ramadan Exposure and children’s
academic performance. Econ. J. 125, 1501–1533. https://doi.org/10.1111/
ecoj.12168.
Amare, M., Jensen, N.D., Shiferaw, B., Cissé, J.D., 2018. Rainfall shocks and agricultural
productivity: implication for rural household consumption. Agric. Syst. 166, 79–89.
Andalon, M., Azevedo, J.P., Rodriguez-Castelan, C., Sanfelice, V., ValderramaGonzalez, D., 2016. Weather shocks and health at birth in colombia. World Dev. 82,
69–82. https://doi.org/10.1016/j.worlddev.2016.01.015.
Andersen, S.L., 2003. Trajectories of brain development: point of vulnerability or
window of opportunity? Neurosci. Biobehav. Rev. 27 (1–2), 3–18. https://doi.org/
10.1016/s0149-7634(03)00005-8.
Austin, M.P., Hadzi-Pavlovic, D., Leader, D., Saint, K., Parker, G., 2005. Maternal trait
anxiety, depression and life event stress in pregnancy: relationships with infant
temperament. Early Hum. Dev. 81 (2), 183–190.
Baez, J.E., Caruso, G., Niu, C., 2017. Tracing Back the Weather Origins of Human
Welfare. Evidence from Mozambique. In: Policy Research Working Paper, No. 8167.
World Bank Group.
Bandura, A., 1993. Perceived self-efficacy in cognitive development and functioning.
Educ. Psychol. 28, 117–148. https://doi.org/10.1207/s15326985ep2802_3.
Banerjee, A., Duflo, E., Postel-Vinay, G., Watts, T., 2010. Long-run health impacts of
income shocks: wine and phylloxera in nineteenth-century France. Rev. Econ. Stat.
92 (4), 714–728. https://doi.org/10.1162/REST_a_00024.
Barker, D.J., 1990. The fetal and infant origins of adult disease. Br. Med. J. 301, 1111.
https://doi.org/10.1136/bmj.301.6761.1111.
Barker, D.J., 1998. In utero programming of chronic disease. Clin. Sci. 95 (2), 115–128.
https://doi.org/10.1042/cs0950115.
Bharadwaj, P., Eberhard, J.P., Neilson, C., 2018. Health at birth, parental investments,
and academic outcomes. J. Labor Econ. 36 (2), 349–394. 〈https://EconPapers.repec.
org/RePEc:ucp:jlabec:doi:10.1086/695616〉.
Bhavani, P., Chakravarthi, V., Roy, P.S., Joshi, P.K., Chandrasekar, K., 2017. “Long-term
agricultural performance and climate variability for drought assessment: a regional
study from telangana and Andhra Pradesh States, India.” Geomatics, natural hazards
and risk, 8 (2), 822–840.
Block, J., 1995. A contrarian view of the five-factor approach to personality description.
Psychol. Bull. 117 (2), 187–215. https://doi.org/10.1037/0033-2909.117.2.187.
Borghans, L., Duckworth, A.L., Heckman, J.J., ter Weel, B., 2008. The economic and
psychology of personality traits. J. Human Resour. 972–1059. https://doi.org/
10.3368/jhr.43.4.972.
Bundervoet, T., Fransen, S., 2018. The educational impact of shocks in utero: evidence
from Rwanda. Econ. Hum. Biol. 29, 88–101. https://doi.org/10.1016/j.
ehb.2018.01.005.
Cameron, L., Shah, M., 2015. Risk taking behaviour in the wake of natural disasters.
J. Hum. Resour. 50 (2), 2484–2515. https://doi.org/10.3368/jhr.50.2.484.
Carrillo, B., 2020. Early Rainfall Shocks and Later-Life Outcomes: Evidence from
Colombia. World Bank Economic Review. World Bank Group,, pp. 179–209.
Cueto, S., Leon J., 2012. Psychometric characteristics of cognitive development and
achievement instruments in round 3 of Young Lives. Young Lives Technical Note No.
25.
Cueto, S., Leon J., Guerrero G., Muñoz I. , 2009. Psychometric characteristics of cognitive
development and achievement instruments in round 2 of Young Lives. Young Lives
Technical Note No. 15.
Cunha, F., Heckman, J., 2008. Formulating, identifying and estimating the technology of
cognitive and noncognitive skill formation. J. Hum. Resour. 43 (4), 738–782.
https://doi.org/10.3368/jhr.43.4.738.
Cunha, F., Heckman, J.J., Schennach, S.M., 2010. Estimating the technology of cognitive
and noncognitive skill formation. Econometrica 78 (3), 883–931. https://doi.org/
10.3982/ECTA6551.
Datar, A., Liu, J., Linnemayr, S., Stecher, C., 2013. The impact of natural disasters on
child health and investments in rural India. Soc. Sci. Med 76 (1), 83–91. https://doi.
org/10.1016/j.socscimed.2012.10.008.
De, U.S., Dube, R.K., Prakasa Rao, G.S., 2005. Extreme Weather Events over India in the
past 100 Years. J. Indian Geophys. Union 9 (3), 173–187.
14
G. Chang et al.
Economics and Human Biology 44 (2022) 101089
Schwarzer, R., Jerusalem, M., 1995. Generalized self-efficacy scale. In: Weinman, J.,
Wright, S., Johnston, M. (Eds.), Measures in Health Psychology: A user’s Portfolio.
Causal and Control Beliefs. NFER-NELSON, Windsor, UK, pp. 35–37.
Shah, M., Steinberg, B.M., 2017. Drought of opportunities: contemporaneous and longterm impacts of rainfall shocks on human capital. J. Political Econ. 125, 2. https://
doi.org/10.1086/690828.
Shenkin, S.D., Starr, J.M., Deary, I.J., 2004. Birth weight and cognitive ability in
childhood: a systematic review. Psychol. Bull. 130 (6), 989–1013. https://doi.org/
10.1037/0033-2909.130.6.989.
Skoufias, E., Vinha, K., 2012. Climate variability and child height in rural Mexico. Econ.
Hum. Biol. 10 (1), 54–73. https://doi.org/10.1016/j.ehb.2011.06.001.
Skoufias, E., Vinha, K., 2013. The impacts of climate variability on household welfare in
rural Mexico. Popul. Environ. 34 (3), 370–399. https://doi.org/10.1007/s11111012-0167-3.
Spears, D., 2012. Height and cognitive achievement among Indian children. Econ. Hum.
Biol. 10 (2), 210–219. https://doi.org/10.1016/j.ehb.2011.08.005. Epub 2011 Aug
25. PMID: 21907646.
Stiles, J., Jernigan, T.L., 2010. The basics of brain development. Neuropsychol. Rev. 20
(4), 327–348. https://doi.org/10.1007/s11065-010-9148-4.
Thai, T.Q., Falaris, M.E., 2014. Child schooling, child health, and rainfall shocks:
evidence from rural Vietnam. J. Dev. Stud. 50 (7), 1025–1037. https://doi.org/
10.1080/00220388.2014.903247.
Thompson, R.A., Nelson, C.A., 2001. Developmental science and the media. Early Brain
Dev. Am. Psychol. 56 (1), 5–15. https://doi.org/10.1037/0003-066X.56.1.5.
Walsh, K., McCormack, C.A., Webster, R., Pinto, A., Lee, S., Feng, T., Krakovsky, H.S.,
O’Grady, S.M., Tycko, B., Champagne, F.A., Werner, E.A., Liu, G., Monk, C., 2019.
Maternal prenatal stress phenotypes associate with fetal neurodevelopment and
birth outcomes. PNAS 116 (48), 23996–24005. https://doi.org/10.1073/
pnas.1905890116.
Webb, D. , 2019. Critical Periods in Cognitive and Socioemotional Development:
Evidence from Weather Shocks in Indonesia. Economics and Finance. ffdumas02407565f.
Yamashita, N., Trinh, T. , 2020. The Effects of Prenatal Exposure to Plentiful Rainfall on
Cognitive Development in Viet Nam. ERIA Discussion Paper, No. 353. Economic
Research Institute for ASEAN and East Asia.
Yorke, L., Ogando Portela, M.J., 2018. Psychosocial scales in the young lives round 4
survey. Selection, Adaptation and Validation. Technical Note. University of Oxford,
Young Lives, p. 45.
Young Lives, 2017. Young Lives Survey Design and Sampling (Round 5); United Andhra
Pradesh. Young Lives Factsheet. October, 2017.
Meier, L.L., Orth, U., Denissen, J.J.A., Kuhnel, A., 2011. Age differences in instability,
contingency, and level of self-esteem across the life span. J. Res. Personal. 45 (6),
604–612. https://doi.org/10.1016/j.jrp.2011.08.008.
, 2004Miguel, E., Kremer, M., 2004. Worms: Identifying impacts on education and health
in the presence of treatment externalities. Econometrica 72 (1), 159–217.
, 1995Mount, M.K., Barrick, M.R., 1995. The big five personality dimensions:
implications for research and practice in human resources management. Res. Pers. Hum.
Resour. Manag. 13, 153–200.
Neelsen S. , Stratmann T. , Early-life famine exposure and later-life outcomes: evidence
from survivors of the greek famine GMU Work. Pap. Econ. 2012 doi: 10.2139/
ssrn.1986723.
Palanisami K. , Haileselassie A. , Reddy K.K. , Ranganathan C.R. , Wani S.P. , Craufurd P. ,
Kumar S. , Climate change, gender and adaptation strategies in dryland systems of
South Asia. A household level analysis In Andhra Pradesh, Karnataka and Rajasthan
States of India Hyderabad: Int. Crops Res. Inst. Semi-Arid Trop. 2015.
, 2015Rocha, R., Soares, R.R., 2015. Water scarcity and birth outcomes in the Brazilian
semiarid. J. Dev. Econ. 112, 72–91. https://doi.org/10.1016/j.jdeveco.2014.10.003.
Rooij, S.R., Wouters, H., Yonker, J.E., Painter, R.C., Roseboom, T.J., 2010. Prenatal
undernutrition and cognitive function in late adulthood. Proc. Natl. Acad. Sci. USA
107 (39) https://doi.org/16881–16886. 10.1073/pnas.1009459107.
Rosales, M.F., 2014. Impact of Early Life Shocks on Human Capital Formation: El Nino
Floods in Ecuador. IDB Working Paper Series, No. IDB-WP-503. Inter-American
Development Bank (IDB).
Rosales-Rueda, M., 2018. The impact of early life shocks on human capital formation:
evidence from El Niño floods in Ecuador. J. Health Econ. 62, 13–44. https://doi.org/
10.1016/j.jhealeco.2018.07.003.
Rose, E., 1999. Consumption Smoothing and Excess Female Mortality in Rural India. Rev.
Econ. Stat. 81 (1), 41–49.
Rosenberg, M., 1965. Society and the Adolescent Self-Image. Princeton University Press,
Princeton, NJ. https://doi.org/10.1126/science.148.3671.804.
Rosenfeld, C.S., 2015. Sex-Specific placental responses in fetal development.
Endocrinology 156 (10), 3422–3434.
, 1966Rotter, J.B., 1966. Generalized expectancies for internal versus external control of
reinforcement. Psychol. Monogr. Gen. Appl. 80 (1), 1–28. https://doi.org/10.1037/
h0092976.
S. Krutikova H.B. Lilleor Fetal Origins of Personality: Effects of early life circumstances
on adult personality traits CSAE Work. Pap. 2015.
Sanchez, A., 2017. Early-life exposure to weather shocks and human capital
accumulation. Evidence from the Peruvian Highlands. Young Lives Working Paper.
Oxford, Young Lives.
15